Unverified50% confidenceOpinionExact time
Mattick refutes LeCun's claim that 'a perfect world model would make RL unnecessary,' arguing that there is a gap between being able to predict the future and being able to infer the correct action.
1
Sources
50%
Confidence
Long-term
Relevance
10/3/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedLeCun argues that LLMs are fundamentally text prediction systems based on statistical patterns, lacking a true world model, causal reasoning, and persistent memory, with a fundamental gap from human cognition70% similarUnverified主流LLM采用下一词预测目标，模型优化的是语言分布拟合度而非事实准确性，RLHF无法从根本上消除幻觉，完全消除幻觉在现有架构下被普遍认为不可能65% similarUnverifiedMattick proposes a distinction: if data collection is inside the loop, it is model-based RL; if outside the loop, it is a world model; Schmidhuber's original world model paper is essentially model-based reinforcement learning.65% similarUnverifiedThe "engineering controllability" camp, represented by Yann LeCun, believes that existing models are essentially statistical pattern-matching machines, far from the level of autonomy that would require a special governance framework.63% similarUnverifiedYann LeCun and others argue that models need to understand the causal structure and physical laws of the world, advocating world models as a necessary component for building artificial general intelligence.63% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/969699API
curl https://kongchang.com/api/v1/knowledge/claims/969699MCP
get_claim(id=969699)